📦 WarehouseMind AI

WarehouseMind AI is a multi-agent AI system that automates warehouse decision-making using intelligent coordination between specialized agents.

It helps warehouse managers:

Monitor inventory levels Detect staffing risks Access warehouse SOPs and safety policies Generate actionable operational insights

The system uses Google ADK (Agent Development Kit) with Gemini 2.5 Flash and integrates a Model Context Protocol (MCP) tool layer for real-world data execution.

---

🧠 Problem Statement

Warehouse operations involve multiple independent systems:

Inventory tracking Workforce management Policy/SOP lookup

Manually analyzing these systems is:

Slow Error-prone Not scalable

WarehouseMind AI solves this by introducing a coordinated multi-agent system that automatically routes queries to the correct domain expert agent.

---

🤖 AI Agent Architecture

🔹 Coordinator Agent (Main Brain) Routes user queries to correct sub-agent Combines responses into final output *🔹 Sub-Agents

1. Inventory Agent

Detects low stock items Identifies restock risks Uses check_inventory() MCP tool

2. Worker Agent

Monitors shift attendance Detects early departures Uses check_worker_status() tool

3. Knowledge Agent

Retrieves SOPs and warehouse policies Uses search_warehouse_documents() tool

---

⚙️ MCP Tool Layer (Important Concept)

This project uses Model Context Protocol (MCP) to connect AI agents with real backend logic.

📌 server.py Acts as the MCP server Exposes tools to agents Handles tool execution requests

📌 tools.py

Contains actual business logic:

Inventory threshold checks Worker status analysis Document retrieval 🔁 Why MCP is used Decouples AI reasoning from backend logic Makes system modular and scalable Allows tool reuse across agents

---

🔄 System Workflow

User Query -> Coordinator Agent (Gemini 2.5 Flash) -> Routes to Sub-Agent -> Sub-Agent calls MCP Tool -> server.py executes tool -> tools.py runs logic -> Response returned to agent -> Final structured answer → UI

---

🚀 Key Features

• Multi-Agent Orchestration: Intelligent routing between Inventory, Worker, and Knowledge agents.
• Inventory Monitoring: Real-time tracking of stock levels with threshold-based risk analysis.
• Worker Optimization: Automated detection of staffing risks and operational impacts.
• Knowledge Retrieval: Instant access to warehouse SOPs, safety manuals, and compliance policies.
• Actionable Intelligence: Provides specific recommendations and next steps for warehouse managers.

---

🏗 System Architecture

• Frontend: React-based interactive chat interface.
• Backend: FastAPI (Python) serving as the bridge between the UI and the agent framework.
• Agent Brain: Google ADK running on Gemini 2.5 Flash.

---

Agent Responsibilities

Agent	Responsibility	Primary Tool
Coordinator	Routes queries & integrates findings	N/A
Inventory	Monitors stock & detects thresholds	check_inventory()
Worker	Tracks attendance & staffing risks	check_worker_status()
Knowledge	Retrieves SOPs & safety policies	search_warehouse_documents()

---

⚙️ Tech Stack

• Language: Python 3.10+
• AI Framework: Google ADK (Agents)
• LLM: Gemini 2.5 Flash
• API: FastAPI
• Frontend: Vite + React
• MCP Tool execution layer for agent-to-function communication
• ASGI Server: Uvicorn (for running FastAPI backend)

---

🖥️ How to Run Locally

git clone https://github.com/Username/WarehouseMind-AI.git
cd WarehouseMind-AI

1. Backend

uvicorn app.main:app --reload

2. Frontend

cd frontend
npm install
npm run dev

---

Example Queries

Inventory Check: "Which products are currently below threshold and what is the restocking risk?"

Workforce Analysis: "Can you list the workers who left early today and explain the operational impact?"

Policy Retrieval: "What is the safety procedure for hazardous material spills?"